Comparison of CWSI and Ts-Ta-VIs in moisture monitoring of dryland crops (sorghum, maize) based on UAV remote sensing

Abstract

Monitoring agricultural drought using remote sensing data is crucial for precision irrigation in modern agriculture. Utilizing unmanned aerial vehicle (UAV) remote sensing, this study explored the applicability of an empirical crop water stress index (CWSI) based on canopy temperature and three-dimensional drought indices (TDDI) constructed from surface temperature (Ts), air temperature (Ta) and five vegetation indices (VIs) for monitoring the moisture status of dryland crops. Three machine learning algorithms (random forest regression [RFR], support vector regression, and partial least squares regression) were used to compare the performance of the drought indices for vegetation moisture content (VMC) estimation in sorghum and maize. The main results of the study were as follows: (1) Comparative analysis of the drought indices revealed that Ts-Ta-Normalized Difference Vegetation Index (TDDIn) and Ts-Ta-Enhanced Vegetation Index (TDDIe) were more strongly correlated with VMC compared with the other indices. The indices exhibited varying sensitivities to VMC under different irrigation regimes; the strongest correlation observed was for the TDDIe index with maize under the fully irrigated treatment (r=−0.93); (2) Regarding spatial and temporal characteristics, the TDDIn, TDDIe and CWSI indices showed minimal differences over the experimental period, with coefficients of variation were 0.25, 0.18 and 0.24, respectively. All three indices were capable of effectively characterizing the moisture distribution in dryland maize and sorghum crops, but the TDDI indices more accurately monitored the spatial distribution of crop moisture after a rainfall or irrigation event. (3) For prediction of the moisture content of single crops, RFR models based on TDDIn and TDDIe estimated VMC most accurately (R20.7), and the TDDIn-based model predicted VMC with the highest accuracy when considering multiple-crop samples, with R2 and RMSE of 0.62 and 14.26%, respectively. Thus, TDDI proved more effective than the CWSI in estimating crop water content.